New TC stats system & progress

[zodac]

19 minutes ago, BWG said:

Likely the team # since there's a wu count but no points. Should be fixed when he switches for EHW FAT.

Remind me of this case (or add it as an issue on github). I think with the new DB should be able to exclude units with no points a bit easier. But will need someone to test with, so let's try it after the monthly reset. 

[Avacado]

How does the 960 on Windows get a 1.12 multiplier and the 960 on Linux get a 6.85 multiplier??? It seems a bit EXTREME, not in a good way though. 

[tictoc]

33 minutes ago, damric said:

I don't know. I've been folding for 239902. I guess I didn't set something right?

 

Nothing to set.  The EHW FAT is just the monthly folding competition. To enter for prizes and/or individual stats tracking during the competition, you just have to post in the thread with "IN" followed by your folding username.

Edited April 22, 2021 by tictoc

[BWG]

15 minutes ago, Avacado said:

How does the 960 on Windows get a 1.12 multiplier and the 960 on Linux get a 6.85 multiplier??? It seems a bit EXTREME, not in a good way though. 

 

Currently, it depends on the top card folding in the category. He's in nVidia against you, and I'm in Wildcard against Axipher. The Windows points base line on LARS for your card is divided by the Linux points baseline on LARS for his 960 which = the multiplier.

[damric]

9 minutes ago, tictoc said:

 

Nothing to set.  The EHW FAT is just the monthly folding competition. To enter for prizes and/or individual stats tracking during the competition, you just have to post in the thread with "IN" followed by your folding username.

Cool so though I was invited in, and told to fold, apparently I wasn't actually in, lol.

 

I need to tear that rig back down for maintenance again anyway so I'll go ahead and finish my W/U.

[BWG]

Yeah, I just forgot to give him the info after you switched hardware. We'll get you put in for when you're back up.

Edited April 22, 2021 by BWG

[damric]

4 minutes ago, BWG said:

Yeah, I just forgot to give him the info after you switched hardware. We'll get you put in for when you're back up.

No big deal. This was kind of a practice run anyway right? 

[BWG]

Yep, demo to test functionality. Take your time, we have 2 teams, but the stats will be reset a few times too to even it back out.

[Avacado]

1 hour ago, BWG said:

 

Currently, it depends on the top card folding in the category. He's in nVidia against you, and I'm in Wildcard against Axipher. The Windows points base line on LARS for your card is divided by the Linux points baseline on LARS for his 960 which = the multiplier.

So this means if he decides to fold on Linux he will get 6x the points compared to him folding on Windows? I am mathematically challenged. It doesn't make sense. 

 

 

 

 

Ok, so what are we dividing here? At quick glance I don't see any divisible numbers that would yield 6x as opposed to 1x. 

 

The TC is supposed to reward Overclocking right? I will have to see, but unless I can find a work around, I am not able to add any voltage to the M40. Not complaining, but might need an overclockable card. Get ready for my GT710 to enter the competition. 

Edited April 22, 2021 by Avacado

[BWG]

I use this page. It's in The ETF Manual. If it still doesn't make sense, let me know. It's going to be interesting to see how this all shakes out. Another huge part of the demo. 

 

Sorry, this page actually is the one you need.

 

I really need to know when you all see issues, and down time if you could. It will give us all a look at how this points system will look going forward. It's still way better than not having any adjustments the way we used to run it. I agree though, if you cannot oc, it might hurt you a bit, so it's something you can play around with when evaluating the hardware you have. 

 

Me for example, I'm watching my HFM Logs for all of my cards I constantly fold for 24 hours. Gives me data to see which card could be the most competitive. I'm certain adjustments could be made down the line too.

 

All I know is this last reset and a series of clock increases has me finally ahead of @axipher

 

Also, we have a wide variety of hardware between 2 full teams, and the points are almost dead even, so I do believe it's looking promising.

 

Edited April 22, 2021 by BWG

[BWG]

There has been a Lead Change! FPN takes #1 from Foldaholics on the last update.

 

[zodac]

8 hours ago, BWG said:

Currently, it depends on the top card folding in the category. He's in nVidia against you, and I'm in Wildcard against Axipher. The Windows points base line on LARS for your card is divided by the Linux points baseline on LARS for his 960 which = the multiplier.

Was the final? Or did you want me to reset the points and trial a different multiplier? Like basing the multipliers on the best available nVidia/Windows, nVidia/Linux, AMD/Windows, AMD/Linux, full stop? Rather than based on who has that hardware?

I'm planning to reset the points again after the weekend (should have the new DB tables ready then), so I could redo the multipliers and we could see how that runs for a few days?

[BWG]

Let's run this for a bit more.

 

But, for everyone else, we have the multipliers setup to the best hardware folding in the category and per each category.

 

What's being considered is to base all point multipliers off the 3090 for the whole event. So, give us your thoughts on that everyone.

 

It allows a static multiplier for each piece of hardware that could be automated down the line.

[Avacado]

1 hour ago, BWG said:

Let's run this for a bit more.

 

But, for everyone else, we have the multipliers setup to the best hardware folding in the category and per each category.

 

What's being considered is to base all point multipliers off the 3090 for the whole event. So, give us your thoughts on that everyone.

 

It allows a static multiplier for each piece of hardware that could be automated down the line.

No issue with the multiplier being based off of the 3090. But the problem remains, as I mentioned awhile back. There is often a large discrepancy between windows vs linux points (Either win or lin advantage). I would still like to know just what the divisible (Exactly WHICH stats are the multi's being generated from) factors are here so I can formulate better questions for you and Zodac. 

[BWG]

20 minutes ago, Avacado said:

No issue with the multiplier being based off of the 3090. But the problem remains, as I mentioned awhile back. There is often a large discrepancy between windows vs linux points (Either win or lin advantage). I would still like to know just what the divisible (Exactly WHICH stats are the multi's being generated from) factors are here so I can formulate better questions for you and Zodac. 

 

We ask for the O/S, and use the PPD estimate for the O/S the user is folding on specifically. That change was made because of you saying that a while back.

[tictoc]

I think the multiplier should be based on the highest ppd GPU. 

 

Also,  you can get some pretty whacky results when you handicap by OS.  For example, the R9 200/300(aka 290, 390, etc) would get nearly a 2x multiplier if you ran it in Linux.  I can say with 100% confidence that Windows is not twice as fast as Linux.  In reality it is pretty much a wash between the two OS's.  There are similar results with newer GPUs as well.  A 2080S is for sure not 27% better when run in Linux.  With the CUDA tasks it is maybe a few percent better.  You can see this on the newer GPUs like the 3090 which is actually better on Windows according to the LARS db.

 

I'm not sure how we account for those types of inconsistencies without just totally eliminating the OS as a variable in the handicapping. 

Edited April 22, 2021 by tictoc

[zodac]

27 minutes ago, Avacado said:

formulate better questions for you and Zodac. 

 

I'll hold my hands up and say I'm just the code monkey here. I just do what gets decided here.

 

Also, you got my name wrong. It's not a big deal the first time... but probably best it doesn't happen again.  Not threatening anything... but we did just discuss adding the ability to "fine" TC users points if necessary.

[Avacado]

1 minute ago, zodac said:

 

I'll hold my hands up and say I'm just the code monkey here. I just do what gets decided here.

 

Also, you got my name wrong. It's not a big deal the first time... but probably best it doesn't happen again.  Not threatening anything... but we did just discuss adding the ability to "fine" TC users points if necessary.

Sorry, I forgot about the capital Z.

Edited April 22, 2021 by Avacado

[Avacado]

8 minutes ago, tictoc said:

I think the multiplier should be based on the highest ppd GPU. 

 

I'm not sure how we account for those types of inconsistencies without just totally eliminating the OS as a variable in the handicapping. 

This. In a nutshell. Speaking of Nutshell, enjoy. 

 

 

 

[BWG]

 

1 minute ago, Avacado said:

Sorry, I forgot about the capital Z.

 

Hey, this is the page we use: https://folding.lar.systems/gpu_ppd/overall_ranks

[zodac]

  

1 minute ago, Avacado said:

Sorry, I forgot about the capital Z.

 

 

It's been a while, so it's understandable.

 

But not really forgivable. I'll remember.

Edited April 22, 2021 by zodac

[Avacado]

1 minute ago, zodac said:

  

 

 

It's been a while, so it's understandable.

 

But not really forgivable. I'll remember.

S***.

[BWG]

Just now, Avacado said:

S***.

 

Someone's 'doomed'.

[Avacado]

12 hours ago, BWG said:

 

Currently, it depends on the top card folding in the category. He's in nVidia against you, and I'm in Wildcard against Axipher. The Windows points base line on LARS for your card is divided by the Linux points baseline on LARS for his 960 which = the multiplier.

My whole string of comments stemmed from this comment. It was confusing. I DO happen to know about the lars sys page BWG, how do you think I decided on M40s?

 

There isn't much difference in the Win/Lin average for either my M40 OR the 960, which is why I couldn't figure out the 1x/6x computation. 

Edited April 22, 2021 by Avacado

[BWG]

4 minutes ago, Avacado said:

My whole string of comments stemmed from this comment. It was confusing. I DO happen to know about the lars sys page BWG, how do you think I decided on M40s?

 

 

 

Edit: I'll outline the math later today.

Edited April 22, 2021 by BWG